xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision e493bec3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: semantics.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/uaccess.h>
13 #include <linux/bitops.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
17 #include <linux/mm.h>
18 #include <linux/string.h>
19 #include <linux/socket.h>
20 #include <linux/sockios.h>
21 #include <linux/errno.h>
22 #include <linux/in.h>
23 #include <linux/inet.h>
24 #include <linux/inetdevice.h>
25 #include <linux/netdevice.h>
26 #include <linux/if_arp.h>
27 #include <linux/proc_fs.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/netlink.h>
32 #include <linux/hash.h>
33 
34 #include <net/arp.h>
35 #include <net/inet_dscp.h>
36 #include <net/ip.h>
37 #include <net/protocol.h>
38 #include <net/route.h>
39 #include <net/tcp.h>
40 #include <net/sock.h>
41 #include <net/ip_fib.h>
42 #include <net/ip6_fib.h>
43 #include <net/nexthop.h>
44 #include <net/netlink.h>
45 #include <net/rtnh.h>
46 #include <net/lwtunnel.h>
47 #include <net/fib_notifier.h>
48 #include <net/addrconf.h>
49 
50 #include "fib_lookup.h"
51 
52 static DEFINE_SPINLOCK(fib_info_lock);
53 static struct hlist_head *fib_info_hash;
54 static struct hlist_head *fib_info_laddrhash;
55 static unsigned int fib_info_hash_size;
56 static unsigned int fib_info_hash_bits;
57 static unsigned int fib_info_cnt;
58 
59 #define DEVINDEX_HASHBITS 8
60 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
61 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
62 
63 /* for_nexthops and change_nexthops only used when nexthop object
64  * is not set in a fib_info. The logic within can reference fib_nh.
65  */
66 #ifdef CONFIG_IP_ROUTE_MULTIPATH
67 
68 #define for_nexthops(fi) {						\
69 	int nhsel; const struct fib_nh *nh;				\
70 	for (nhsel = 0, nh = (fi)->fib_nh;				\
71 	     nhsel < fib_info_num_path((fi));				\
72 	     nh++, nhsel++)
73 
74 #define change_nexthops(fi) {						\
75 	int nhsel; struct fib_nh *nexthop_nh;				\
76 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
77 	     nhsel < fib_info_num_path((fi));				\
78 	     nexthop_nh++, nhsel++)
79 
80 #else /* CONFIG_IP_ROUTE_MULTIPATH */
81 
82 /* Hope, that gcc will optimize it to get rid of dummy loop */
83 
84 #define for_nexthops(fi) {						\
85 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
86 	for (nhsel = 0; nhsel < 1; nhsel++)
87 
88 #define change_nexthops(fi) {						\
89 	int nhsel;							\
90 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
91 	for (nhsel = 0; nhsel < 1; nhsel++)
92 
93 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
94 
95 #define endfor_nexthops(fi) }
96 
97 
98 const struct fib_prop fib_props[RTN_MAX + 1] = {
99 	[RTN_UNSPEC] = {
100 		.error	= 0,
101 		.scope	= RT_SCOPE_NOWHERE,
102 	},
103 	[RTN_UNICAST] = {
104 		.error	= 0,
105 		.scope	= RT_SCOPE_UNIVERSE,
106 	},
107 	[RTN_LOCAL] = {
108 		.error	= 0,
109 		.scope	= RT_SCOPE_HOST,
110 	},
111 	[RTN_BROADCAST] = {
112 		.error	= 0,
113 		.scope	= RT_SCOPE_LINK,
114 	},
115 	[RTN_ANYCAST] = {
116 		.error	= 0,
117 		.scope	= RT_SCOPE_LINK,
118 	},
119 	[RTN_MULTICAST] = {
120 		.error	= 0,
121 		.scope	= RT_SCOPE_UNIVERSE,
122 	},
123 	[RTN_BLACKHOLE] = {
124 		.error	= -EINVAL,
125 		.scope	= RT_SCOPE_UNIVERSE,
126 	},
127 	[RTN_UNREACHABLE] = {
128 		.error	= -EHOSTUNREACH,
129 		.scope	= RT_SCOPE_UNIVERSE,
130 	},
131 	[RTN_PROHIBIT] = {
132 		.error	= -EACCES,
133 		.scope	= RT_SCOPE_UNIVERSE,
134 	},
135 	[RTN_THROW] = {
136 		.error	= -EAGAIN,
137 		.scope	= RT_SCOPE_UNIVERSE,
138 	},
139 	[RTN_NAT] = {
140 		.error	= -EINVAL,
141 		.scope	= RT_SCOPE_NOWHERE,
142 	},
143 	[RTN_XRESOLVE] = {
144 		.error	= -EINVAL,
145 		.scope	= RT_SCOPE_NOWHERE,
146 	},
147 };
148 
149 static void rt_fibinfo_free(struct rtable __rcu **rtp)
150 {
151 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
152 
153 	if (!rt)
154 		return;
155 
156 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
157 	 * because we waited an RCU grace period before calling
158 	 * free_fib_info_rcu()
159 	 */
160 
161 	dst_dev_put(&rt->dst);
162 	dst_release_immediate(&rt->dst);
163 }
164 
165 static void free_nh_exceptions(struct fib_nh_common *nhc)
166 {
167 	struct fnhe_hash_bucket *hash;
168 	int i;
169 
170 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
171 	if (!hash)
172 		return;
173 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
174 		struct fib_nh_exception *fnhe;
175 
176 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
177 		while (fnhe) {
178 			struct fib_nh_exception *next;
179 
180 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
181 
182 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
183 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
184 
185 			kfree(fnhe);
186 
187 			fnhe = next;
188 		}
189 	}
190 	kfree(hash);
191 }
192 
193 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
194 {
195 	int cpu;
196 
197 	if (!rtp)
198 		return;
199 
200 	for_each_possible_cpu(cpu) {
201 		struct rtable *rt;
202 
203 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
204 		if (rt) {
205 			dst_dev_put(&rt->dst);
206 			dst_release_immediate(&rt->dst);
207 		}
208 	}
209 	free_percpu(rtp);
210 }
211 
212 void fib_nh_common_release(struct fib_nh_common *nhc)
213 {
214 	netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
215 	lwtstate_put(nhc->nhc_lwtstate);
216 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
217 	rt_fibinfo_free(&nhc->nhc_rth_input);
218 	free_nh_exceptions(nhc);
219 }
220 EXPORT_SYMBOL_GPL(fib_nh_common_release);
221 
222 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
223 {
224 #ifdef CONFIG_IP_ROUTE_CLASSID
225 	if (fib_nh->nh_tclassid)
226 		atomic_dec(&net->ipv4.fib_num_tclassid_users);
227 #endif
228 	fib_nh_common_release(&fib_nh->nh_common);
229 }
230 
231 /* Release a nexthop info record */
232 static void free_fib_info_rcu(struct rcu_head *head)
233 {
234 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
235 
236 	if (fi->nh) {
237 		nexthop_put(fi->nh);
238 	} else {
239 		change_nexthops(fi) {
240 			fib_nh_release(fi->fib_net, nexthop_nh);
241 		} endfor_nexthops(fi);
242 	}
243 
244 	ip_fib_metrics_put(fi->fib_metrics);
245 
246 	kfree(fi);
247 }
248 
249 void free_fib_info(struct fib_info *fi)
250 {
251 	if (fi->fib_dead == 0) {
252 		pr_warn("Freeing alive fib_info %p\n", fi);
253 		return;
254 	}
255 
256 	call_rcu(&fi->rcu, free_fib_info_rcu);
257 }
258 EXPORT_SYMBOL_GPL(free_fib_info);
259 
260 void fib_release_info(struct fib_info *fi)
261 {
262 	spin_lock_bh(&fib_info_lock);
263 	if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
264 		hlist_del(&fi->fib_hash);
265 
266 		/* Paired with READ_ONCE() in fib_create_info(). */
267 		WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
268 
269 		if (fi->fib_prefsrc)
270 			hlist_del(&fi->fib_lhash);
271 		if (fi->nh) {
272 			list_del(&fi->nh_list);
273 		} else {
274 			change_nexthops(fi) {
275 				if (!nexthop_nh->fib_nh_dev)
276 					continue;
277 				hlist_del(&nexthop_nh->nh_hash);
278 			} endfor_nexthops(fi)
279 		}
280 		fi->fib_dead = 1;
281 		fib_info_put(fi);
282 	}
283 	spin_unlock_bh(&fib_info_lock);
284 }
285 
286 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
287 {
288 	const struct fib_nh *onh;
289 
290 	if (fi->nh || ofi->nh)
291 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
292 
293 	if (ofi->fib_nhs == 0)
294 		return 0;
295 
296 	for_nexthops(fi) {
297 		onh = fib_info_nh(ofi, nhsel);
298 
299 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
300 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
301 		    nh->fib_nh_scope != onh->fib_nh_scope ||
302 #ifdef CONFIG_IP_ROUTE_MULTIPATH
303 		    nh->fib_nh_weight != onh->fib_nh_weight ||
304 #endif
305 #ifdef CONFIG_IP_ROUTE_CLASSID
306 		    nh->nh_tclassid != onh->nh_tclassid ||
307 #endif
308 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
309 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
310 			return -1;
311 
312 		if (nh->fib_nh_gw_family == AF_INET &&
313 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
314 			return -1;
315 
316 		if (nh->fib_nh_gw_family == AF_INET6 &&
317 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
318 			return -1;
319 	} endfor_nexthops(fi);
320 	return 0;
321 }
322 
323 static inline unsigned int fib_devindex_hashfn(unsigned int val)
324 {
325 	return hash_32(val, DEVINDEX_HASHBITS);
326 }
327 
328 static struct hlist_head *
329 fib_info_devhash_bucket(const struct net_device *dev)
330 {
331 	u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
332 
333 	return &fib_info_devhash[fib_devindex_hashfn(val)];
334 }
335 
336 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
337 				      u32 prefsrc, u32 priority)
338 {
339 	unsigned int val = init_val;
340 
341 	val ^= (protocol << 8) | scope;
342 	val ^= prefsrc;
343 	val ^= priority;
344 
345 	return val;
346 }
347 
348 static unsigned int fib_info_hashfn_result(unsigned int val)
349 {
350 	unsigned int mask = (fib_info_hash_size - 1);
351 
352 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
353 }
354 
355 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
356 {
357 	unsigned int val;
358 
359 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
360 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
361 				fi->fib_priority);
362 
363 	if (fi->nh) {
364 		val ^= fib_devindex_hashfn(fi->nh->id);
365 	} else {
366 		for_nexthops(fi) {
367 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
368 		} endfor_nexthops(fi)
369 	}
370 
371 	return fib_info_hashfn_result(val);
372 }
373 
374 /* no metrics, only nexthop id */
375 static struct fib_info *fib_find_info_nh(struct net *net,
376 					 const struct fib_config *cfg)
377 {
378 	struct hlist_head *head;
379 	struct fib_info *fi;
380 	unsigned int hash;
381 
382 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
383 				 cfg->fc_protocol, cfg->fc_scope,
384 				 (__force u32)cfg->fc_prefsrc,
385 				 cfg->fc_priority);
386 	hash = fib_info_hashfn_result(hash);
387 	head = &fib_info_hash[hash];
388 
389 	hlist_for_each_entry(fi, head, fib_hash) {
390 		if (!net_eq(fi->fib_net, net))
391 			continue;
392 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
393 			continue;
394 		if (cfg->fc_protocol == fi->fib_protocol &&
395 		    cfg->fc_scope == fi->fib_scope &&
396 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
397 		    cfg->fc_priority == fi->fib_priority &&
398 		    cfg->fc_type == fi->fib_type &&
399 		    cfg->fc_table == fi->fib_tb_id &&
400 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
401 			return fi;
402 	}
403 
404 	return NULL;
405 }
406 
407 static struct fib_info *fib_find_info(struct fib_info *nfi)
408 {
409 	struct hlist_head *head;
410 	struct fib_info *fi;
411 	unsigned int hash;
412 
413 	hash = fib_info_hashfn(nfi);
414 	head = &fib_info_hash[hash];
415 
416 	hlist_for_each_entry(fi, head, fib_hash) {
417 		if (!net_eq(fi->fib_net, nfi->fib_net))
418 			continue;
419 		if (fi->fib_nhs != nfi->fib_nhs)
420 			continue;
421 		if (nfi->fib_protocol == fi->fib_protocol &&
422 		    nfi->fib_scope == fi->fib_scope &&
423 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
424 		    nfi->fib_priority == fi->fib_priority &&
425 		    nfi->fib_type == fi->fib_type &&
426 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
427 			   sizeof(u32) * RTAX_MAX) == 0 &&
428 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
429 		    nh_comp(fi, nfi) == 0)
430 			return fi;
431 	}
432 
433 	return NULL;
434 }
435 
436 /* Check, that the gateway is already configured.
437  * Used only by redirect accept routine.
438  */
439 int ip_fib_check_default(__be32 gw, struct net_device *dev)
440 {
441 	struct hlist_head *head;
442 	struct fib_nh *nh;
443 
444 	spin_lock(&fib_info_lock);
445 
446 	head = fib_info_devhash_bucket(dev);
447 
448 	hlist_for_each_entry(nh, head, nh_hash) {
449 		if (nh->fib_nh_dev == dev &&
450 		    nh->fib_nh_gw4 == gw &&
451 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
452 			spin_unlock(&fib_info_lock);
453 			return 0;
454 		}
455 	}
456 
457 	spin_unlock(&fib_info_lock);
458 
459 	return -1;
460 }
461 
462 size_t fib_nlmsg_size(struct fib_info *fi)
463 {
464 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
465 			 + nla_total_size(4) /* RTA_TABLE */
466 			 + nla_total_size(4) /* RTA_DST */
467 			 + nla_total_size(4) /* RTA_PRIORITY */
468 			 + nla_total_size(4) /* RTA_PREFSRC */
469 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
470 	unsigned int nhs = fib_info_num_path(fi);
471 
472 	/* space for nested metrics */
473 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
474 
475 	if (fi->nh)
476 		payload += nla_total_size(4); /* RTA_NH_ID */
477 
478 	if (nhs) {
479 		size_t nh_encapsize = 0;
480 		/* Also handles the special case nhs == 1 */
481 
482 		/* each nexthop is packed in an attribute */
483 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
484 		unsigned int i;
485 
486 		/* may contain flow and gateway attribute */
487 		nhsize += 2 * nla_total_size(4);
488 
489 		/* grab encap info */
490 		for (i = 0; i < fib_info_num_path(fi); i++) {
491 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
492 
493 			if (nhc->nhc_lwtstate) {
494 				/* RTA_ENCAP_TYPE */
495 				nh_encapsize += lwtunnel_get_encap_size(
496 						nhc->nhc_lwtstate);
497 				/* RTA_ENCAP */
498 				nh_encapsize +=  nla_total_size(2);
499 			}
500 		}
501 
502 		/* all nexthops are packed in a nested attribute */
503 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
504 
505 	}
506 
507 	return payload;
508 }
509 
510 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
511 	       int dst_len, u32 tb_id, const struct nl_info *info,
512 	       unsigned int nlm_flags)
513 {
514 	struct fib_rt_info fri;
515 	struct sk_buff *skb;
516 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
517 	int err = -ENOBUFS;
518 
519 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
520 	if (!skb)
521 		goto errout;
522 
523 	fri.fi = fa->fa_info;
524 	fri.tb_id = tb_id;
525 	fri.dst = key;
526 	fri.dst_len = dst_len;
527 	fri.dscp = fa->fa_dscp;
528 	fri.type = fa->fa_type;
529 	fri.offload = READ_ONCE(fa->offload);
530 	fri.trap = READ_ONCE(fa->trap);
531 	fri.offload_failed = READ_ONCE(fa->offload_failed);
532 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
533 	if (err < 0) {
534 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
535 		WARN_ON(err == -EMSGSIZE);
536 		kfree_skb(skb);
537 		goto errout;
538 	}
539 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
540 		    info->nlh, GFP_KERNEL);
541 	return;
542 errout:
543 	if (err < 0)
544 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
545 }
546 
547 static int fib_detect_death(struct fib_info *fi, int order,
548 			    struct fib_info **last_resort, int *last_idx,
549 			    int dflt)
550 {
551 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
552 	struct neighbour *n;
553 	int state = NUD_NONE;
554 
555 	if (likely(nhc->nhc_gw_family == AF_INET))
556 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
557 	else if (nhc->nhc_gw_family == AF_INET6)
558 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
559 				 nhc->nhc_dev);
560 	else
561 		n = NULL;
562 
563 	if (n) {
564 		state = n->nud_state;
565 		neigh_release(n);
566 	} else {
567 		return 0;
568 	}
569 	if (state == NUD_REACHABLE)
570 		return 0;
571 	if ((state & NUD_VALID) && order != dflt)
572 		return 0;
573 	if ((state & NUD_VALID) ||
574 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
575 		*last_resort = fi;
576 		*last_idx = order;
577 	}
578 	return 1;
579 }
580 
581 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
582 		       struct nlattr *encap, u16 encap_type,
583 		       void *cfg, gfp_t gfp_flags,
584 		       struct netlink_ext_ack *extack)
585 {
586 	int err;
587 
588 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
589 						    gfp_flags);
590 	if (!nhc->nhc_pcpu_rth_output)
591 		return -ENOMEM;
592 
593 	if (encap) {
594 		struct lwtunnel_state *lwtstate;
595 
596 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
597 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
598 			err = -EINVAL;
599 			goto lwt_failure;
600 		}
601 		err = lwtunnel_build_state(net, encap_type, encap,
602 					   nhc->nhc_family, cfg, &lwtstate,
603 					   extack);
604 		if (err)
605 			goto lwt_failure;
606 
607 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
608 	}
609 
610 	return 0;
611 
612 lwt_failure:
613 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
614 	nhc->nhc_pcpu_rth_output = NULL;
615 	return err;
616 }
617 EXPORT_SYMBOL_GPL(fib_nh_common_init);
618 
619 int fib_nh_init(struct net *net, struct fib_nh *nh,
620 		struct fib_config *cfg, int nh_weight,
621 		struct netlink_ext_ack *extack)
622 {
623 	int err;
624 
625 	nh->fib_nh_family = AF_INET;
626 
627 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
628 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
629 	if (err)
630 		return err;
631 
632 	nh->fib_nh_oif = cfg->fc_oif;
633 	nh->fib_nh_gw_family = cfg->fc_gw_family;
634 	if (cfg->fc_gw_family == AF_INET)
635 		nh->fib_nh_gw4 = cfg->fc_gw4;
636 	else if (cfg->fc_gw_family == AF_INET6)
637 		nh->fib_nh_gw6 = cfg->fc_gw6;
638 
639 	nh->fib_nh_flags = cfg->fc_flags;
640 
641 #ifdef CONFIG_IP_ROUTE_CLASSID
642 	nh->nh_tclassid = cfg->fc_flow;
643 	if (nh->nh_tclassid)
644 		atomic_inc(&net->ipv4.fib_num_tclassid_users);
645 #endif
646 #ifdef CONFIG_IP_ROUTE_MULTIPATH
647 	nh->fib_nh_weight = nh_weight;
648 #endif
649 	return 0;
650 }
651 
652 #ifdef CONFIG_IP_ROUTE_MULTIPATH
653 
654 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
655 			      struct netlink_ext_ack *extack)
656 {
657 	int nhs = 0;
658 
659 	while (rtnh_ok(rtnh, remaining)) {
660 		nhs++;
661 		rtnh = rtnh_next(rtnh, &remaining);
662 	}
663 
664 	/* leftover implies invalid nexthop configuration, discard it */
665 	if (remaining > 0) {
666 		NL_SET_ERR_MSG(extack,
667 			       "Invalid nexthop configuration - extra data after nexthops");
668 		nhs = 0;
669 	}
670 
671 	return nhs;
672 }
673 
674 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
675 			    struct netlink_ext_ack *extack)
676 {
677 	if (nla_len(nla) < sizeof(*gw)) {
678 		NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
679 		return -EINVAL;
680 	}
681 
682 	*gw = nla_get_in_addr(nla);
683 
684 	return 0;
685 }
686 
687 /* only called when fib_nh is integrated into fib_info */
688 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
689 		       int remaining, struct fib_config *cfg,
690 		       struct netlink_ext_ack *extack)
691 {
692 	struct net *net = fi->fib_net;
693 	struct fib_config fib_cfg;
694 	struct fib_nh *nh;
695 	int ret;
696 
697 	change_nexthops(fi) {
698 		int attrlen;
699 
700 		memset(&fib_cfg, 0, sizeof(fib_cfg));
701 
702 		if (!rtnh_ok(rtnh, remaining)) {
703 			NL_SET_ERR_MSG(extack,
704 				       "Invalid nexthop configuration - extra data after nexthop");
705 			return -EINVAL;
706 		}
707 
708 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
709 			NL_SET_ERR_MSG(extack,
710 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
711 			return -EINVAL;
712 		}
713 
714 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
715 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
716 
717 		attrlen = rtnh_attrlen(rtnh);
718 		if (attrlen > 0) {
719 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
720 
721 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
722 			nlav = nla_find(attrs, attrlen, RTA_VIA);
723 			if (nla && nlav) {
724 				NL_SET_ERR_MSG(extack,
725 					       "Nexthop configuration can not contain both GATEWAY and VIA");
726 				return -EINVAL;
727 			}
728 			if (nla) {
729 				ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
730 						       extack);
731 				if (ret)
732 					goto errout;
733 
734 				if (fib_cfg.fc_gw4)
735 					fib_cfg.fc_gw_family = AF_INET;
736 			} else if (nlav) {
737 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
738 				if (ret)
739 					goto errout;
740 			}
741 
742 			nla = nla_find(attrs, attrlen, RTA_FLOW);
743 			if (nla) {
744 				if (nla_len(nla) < sizeof(u32)) {
745 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
746 					return -EINVAL;
747 				}
748 				fib_cfg.fc_flow = nla_get_u32(nla);
749 			}
750 
751 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
752 			/* RTA_ENCAP_TYPE length checked in
753 			 * lwtunnel_valid_encap_type_attr
754 			 */
755 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
756 			if (nla)
757 				fib_cfg.fc_encap_type = nla_get_u16(nla);
758 		}
759 
760 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
761 				  rtnh->rtnh_hops + 1, extack);
762 		if (ret)
763 			goto errout;
764 
765 		rtnh = rtnh_next(rtnh, &remaining);
766 	} endfor_nexthops(fi);
767 
768 	ret = -EINVAL;
769 	nh = fib_info_nh(fi, 0);
770 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
771 		NL_SET_ERR_MSG(extack,
772 			       "Nexthop device index does not match RTA_OIF");
773 		goto errout;
774 	}
775 	if (cfg->fc_gw_family) {
776 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
777 		    (cfg->fc_gw_family == AF_INET &&
778 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
779 		    (cfg->fc_gw_family == AF_INET6 &&
780 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
781 			NL_SET_ERR_MSG(extack,
782 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
783 			goto errout;
784 		}
785 	}
786 #ifdef CONFIG_IP_ROUTE_CLASSID
787 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
788 		NL_SET_ERR_MSG(extack,
789 			       "Nexthop class id does not match RTA_FLOW");
790 		goto errout;
791 	}
792 #endif
793 	ret = 0;
794 errout:
795 	return ret;
796 }
797 
798 /* only called when fib_nh is integrated into fib_info */
799 static void fib_rebalance(struct fib_info *fi)
800 {
801 	int total;
802 	int w;
803 
804 	if (fib_info_num_path(fi) < 2)
805 		return;
806 
807 	total = 0;
808 	for_nexthops(fi) {
809 		if (nh->fib_nh_flags & RTNH_F_DEAD)
810 			continue;
811 
812 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
813 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
814 			continue;
815 
816 		total += nh->fib_nh_weight;
817 	} endfor_nexthops(fi);
818 
819 	w = 0;
820 	change_nexthops(fi) {
821 		int upper_bound;
822 
823 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
824 			upper_bound = -1;
825 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
826 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
827 			upper_bound = -1;
828 		} else {
829 			w += nexthop_nh->fib_nh_weight;
830 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
831 							    total) - 1;
832 		}
833 
834 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
835 	} endfor_nexthops(fi);
836 }
837 #else /* CONFIG_IP_ROUTE_MULTIPATH */
838 
839 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
840 		       int remaining, struct fib_config *cfg,
841 		       struct netlink_ext_ack *extack)
842 {
843 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
844 
845 	return -EINVAL;
846 }
847 
848 #define fib_rebalance(fi) do { } while (0)
849 
850 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
851 
852 static int fib_encap_match(struct net *net, u16 encap_type,
853 			   struct nlattr *encap,
854 			   const struct fib_nh *nh,
855 			   const struct fib_config *cfg,
856 			   struct netlink_ext_ack *extack)
857 {
858 	struct lwtunnel_state *lwtstate;
859 	int ret, result = 0;
860 
861 	if (encap_type == LWTUNNEL_ENCAP_NONE)
862 		return 0;
863 
864 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
865 				   cfg, &lwtstate, extack);
866 	if (!ret) {
867 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
868 		lwtstate_free(lwtstate);
869 	}
870 
871 	return result;
872 }
873 
874 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
875 		 struct netlink_ext_ack *extack)
876 {
877 #ifdef CONFIG_IP_ROUTE_MULTIPATH
878 	struct rtnexthop *rtnh;
879 	int remaining;
880 #endif
881 
882 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
883 		return 1;
884 
885 	if (cfg->fc_nh_id) {
886 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
887 			return 0;
888 		return 1;
889 	}
890 
891 	if (fi->nh) {
892 		if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
893 			return 1;
894 		return 0;
895 	}
896 
897 	if (cfg->fc_oif || cfg->fc_gw_family) {
898 		struct fib_nh *nh;
899 
900 		nh = fib_info_nh(fi, 0);
901 		if (cfg->fc_encap) {
902 			if (fib_encap_match(net, cfg->fc_encap_type,
903 					    cfg->fc_encap, nh, cfg, extack))
904 				return 1;
905 		}
906 #ifdef CONFIG_IP_ROUTE_CLASSID
907 		if (cfg->fc_flow &&
908 		    cfg->fc_flow != nh->nh_tclassid)
909 			return 1;
910 #endif
911 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
912 		    (cfg->fc_gw_family &&
913 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
914 			return 1;
915 
916 		if (cfg->fc_gw_family == AF_INET &&
917 		    cfg->fc_gw4 != nh->fib_nh_gw4)
918 			return 1;
919 
920 		if (cfg->fc_gw_family == AF_INET6 &&
921 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
922 			return 1;
923 
924 		return 0;
925 	}
926 
927 #ifdef CONFIG_IP_ROUTE_MULTIPATH
928 	if (!cfg->fc_mp)
929 		return 0;
930 
931 	rtnh = cfg->fc_mp;
932 	remaining = cfg->fc_mp_len;
933 
934 	for_nexthops(fi) {
935 		int attrlen;
936 
937 		if (!rtnh_ok(rtnh, remaining))
938 			return -EINVAL;
939 
940 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
941 			return 1;
942 
943 		attrlen = rtnh_attrlen(rtnh);
944 		if (attrlen > 0) {
945 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
946 			int err;
947 
948 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
949 			nlav = nla_find(attrs, attrlen, RTA_VIA);
950 			if (nla && nlav) {
951 				NL_SET_ERR_MSG(extack,
952 					       "Nexthop configuration can not contain both GATEWAY and VIA");
953 				return -EINVAL;
954 			}
955 
956 			if (nla) {
957 				__be32 gw;
958 
959 				err = fib_gw_from_attr(&gw, nla, extack);
960 				if (err)
961 					return err;
962 
963 				if (nh->fib_nh_gw_family != AF_INET ||
964 				    gw != nh->fib_nh_gw4)
965 					return 1;
966 			} else if (nlav) {
967 				struct fib_config cfg2;
968 
969 				err = fib_gw_from_via(&cfg2, nlav, extack);
970 				if (err)
971 					return err;
972 
973 				switch (nh->fib_nh_gw_family) {
974 				case AF_INET:
975 					if (cfg2.fc_gw_family != AF_INET ||
976 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
977 						return 1;
978 					break;
979 				case AF_INET6:
980 					if (cfg2.fc_gw_family != AF_INET6 ||
981 					    ipv6_addr_cmp(&cfg2.fc_gw6,
982 							  &nh->fib_nh_gw6))
983 						return 1;
984 					break;
985 				}
986 			}
987 
988 #ifdef CONFIG_IP_ROUTE_CLASSID
989 			nla = nla_find(attrs, attrlen, RTA_FLOW);
990 			if (nla) {
991 				if (nla_len(nla) < sizeof(u32)) {
992 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
993 					return -EINVAL;
994 				}
995 				if (nla_get_u32(nla) != nh->nh_tclassid)
996 					return 1;
997 			}
998 #endif
999 		}
1000 
1001 		rtnh = rtnh_next(rtnh, &remaining);
1002 	} endfor_nexthops(fi);
1003 #endif
1004 	return 0;
1005 }
1006 
1007 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1008 {
1009 	struct nlattr *nla;
1010 	int remaining;
1011 
1012 	if (!cfg->fc_mx)
1013 		return true;
1014 
1015 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1016 		int type = nla_type(nla);
1017 		u32 fi_val, val;
1018 
1019 		if (!type)
1020 			continue;
1021 		if (type > RTAX_MAX)
1022 			return false;
1023 
1024 		if (type == RTAX_CC_ALGO) {
1025 			char tmp[TCP_CA_NAME_MAX];
1026 			bool ecn_ca = false;
1027 
1028 			nla_strscpy(tmp, nla, sizeof(tmp));
1029 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1030 		} else {
1031 			if (nla_len(nla) != sizeof(u32))
1032 				return false;
1033 			val = nla_get_u32(nla);
1034 		}
1035 
1036 		fi_val = fi->fib_metrics->metrics[type - 1];
1037 		if (type == RTAX_FEATURES)
1038 			fi_val &= ~DST_FEATURE_ECN_CA;
1039 
1040 		if (fi_val != val)
1041 			return false;
1042 	}
1043 
1044 	return true;
1045 }
1046 
1047 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1048 			      u32 table, struct netlink_ext_ack *extack)
1049 {
1050 	struct fib6_config cfg = {
1051 		.fc_table = table,
1052 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1053 		.fc_ifindex = nh->fib_nh_oif,
1054 		.fc_gateway = nh->fib_nh_gw6,
1055 	};
1056 	struct fib6_nh fib6_nh = {};
1057 	int err;
1058 
1059 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1060 	if (!err) {
1061 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1062 		netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1063 			    GFP_KERNEL);
1064 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1065 		nh->fib_nh_scope = RT_SCOPE_LINK;
1066 
1067 		ipv6_stub->fib6_nh_release(&fib6_nh);
1068 	}
1069 
1070 	return err;
1071 }
1072 
1073 /*
1074  * Picture
1075  * -------
1076  *
1077  * Semantics of nexthop is very messy by historical reasons.
1078  * We have to take into account, that:
1079  * a) gateway can be actually local interface address,
1080  *    so that gatewayed route is direct.
1081  * b) gateway must be on-link address, possibly
1082  *    described not by an ifaddr, but also by a direct route.
1083  * c) If both gateway and interface are specified, they should not
1084  *    contradict.
1085  * d) If we use tunnel routes, gateway could be not on-link.
1086  *
1087  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1088  * results in pretty ugly and hairy code with obscure logic.
1089  *
1090  * I chose to generalized it instead, so that the size
1091  * of code does not increase practically, but it becomes
1092  * much more general.
1093  * Every prefix is assigned a "scope" value: "host" is local address,
1094  * "link" is direct route,
1095  * [ ... "site" ... "interior" ... ]
1096  * and "universe" is true gateway route with global meaning.
1097  *
1098  * Every prefix refers to a set of "nexthop"s (gw, oif),
1099  * where gw must have narrower scope. This recursion stops
1100  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1101  * which means that gw is forced to be on link.
1102  *
1103  * Code is still hairy, but now it is apparently logically
1104  * consistent and very flexible. F.e. as by-product it allows
1105  * to co-exists in peace independent exterior and interior
1106  * routing processes.
1107  *
1108  * Normally it looks as following.
1109  *
1110  * {universe prefix}  -> (gw, oif) [scope link]
1111  *		  |
1112  *		  |-> {link prefix} -> (gw, oif) [scope local]
1113  *					|
1114  *					|-> {local prefix} (terminal node)
1115  */
1116 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1117 			      u8 scope, struct netlink_ext_ack *extack)
1118 {
1119 	struct net_device *dev;
1120 	struct fib_result res;
1121 	int err = 0;
1122 
1123 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1124 		unsigned int addr_type;
1125 
1126 		if (scope >= RT_SCOPE_LINK) {
1127 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1128 			return -EINVAL;
1129 		}
1130 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1131 		if (!dev) {
1132 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1133 			return -ENODEV;
1134 		}
1135 		if (!(dev->flags & IFF_UP)) {
1136 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1137 			return -ENETDOWN;
1138 		}
1139 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1140 		if (addr_type != RTN_UNICAST) {
1141 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1142 			return -EINVAL;
1143 		}
1144 		if (!netif_carrier_ok(dev))
1145 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1146 		nh->fib_nh_dev = dev;
1147 		netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1148 		nh->fib_nh_scope = RT_SCOPE_LINK;
1149 		return 0;
1150 	}
1151 	rcu_read_lock();
1152 	{
1153 		struct fib_table *tbl = NULL;
1154 		struct flowi4 fl4 = {
1155 			.daddr = nh->fib_nh_gw4,
1156 			.flowi4_scope = scope + 1,
1157 			.flowi4_oif = nh->fib_nh_oif,
1158 			.flowi4_iif = LOOPBACK_IFINDEX,
1159 		};
1160 
1161 		/* It is not necessary, but requires a bit of thinking */
1162 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1163 			fl4.flowi4_scope = RT_SCOPE_LINK;
1164 
1165 		if (table && table != RT_TABLE_MAIN)
1166 			tbl = fib_get_table(net, table);
1167 
1168 		if (tbl)
1169 			err = fib_table_lookup(tbl, &fl4, &res,
1170 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1171 					       FIB_LOOKUP_NOREF);
1172 
1173 		/* on error or if no table given do full lookup. This
1174 		 * is needed for example when nexthops are in the local
1175 		 * table rather than the given table
1176 		 */
1177 		if (!tbl || err) {
1178 			err = fib_lookup(net, &fl4, &res,
1179 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1180 		}
1181 
1182 		if (err) {
1183 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1184 			goto out;
1185 		}
1186 	}
1187 
1188 	err = -EINVAL;
1189 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1190 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1191 		goto out;
1192 	}
1193 	nh->fib_nh_scope = res.scope;
1194 	nh->fib_nh_oif = FIB_RES_OIF(res);
1195 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1196 	if (!dev) {
1197 		NL_SET_ERR_MSG(extack,
1198 			       "No egress device for nexthop gateway");
1199 		goto out;
1200 	}
1201 	netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1202 	if (!netif_carrier_ok(dev))
1203 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1204 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1205 out:
1206 	rcu_read_unlock();
1207 	return err;
1208 }
1209 
1210 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1211 			      struct netlink_ext_ack *extack)
1212 {
1213 	struct in_device *in_dev;
1214 	int err;
1215 
1216 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1217 		NL_SET_ERR_MSG(extack,
1218 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1219 		return -EINVAL;
1220 	}
1221 
1222 	rcu_read_lock();
1223 
1224 	err = -ENODEV;
1225 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1226 	if (!in_dev)
1227 		goto out;
1228 	err = -ENETDOWN;
1229 	if (!(in_dev->dev->flags & IFF_UP)) {
1230 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1231 		goto out;
1232 	}
1233 
1234 	nh->fib_nh_dev = in_dev->dev;
1235 	netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1236 	nh->fib_nh_scope = RT_SCOPE_HOST;
1237 	if (!netif_carrier_ok(nh->fib_nh_dev))
1238 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1239 	err = 0;
1240 out:
1241 	rcu_read_unlock();
1242 	return err;
1243 }
1244 
1245 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1246 		 struct netlink_ext_ack *extack)
1247 {
1248 	int err;
1249 
1250 	if (nh->fib_nh_gw_family == AF_INET)
1251 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1252 	else if (nh->fib_nh_gw_family == AF_INET6)
1253 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1254 	else
1255 		err = fib_check_nh_nongw(net, nh, extack);
1256 
1257 	return err;
1258 }
1259 
1260 static struct hlist_head *
1261 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1262 {
1263 	u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1264 			   fib_info_hash_bits);
1265 
1266 	return &fib_info_laddrhash[slot];
1267 }
1268 
1269 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1270 			       struct hlist_head *new_laddrhash,
1271 			       unsigned int new_size)
1272 {
1273 	struct hlist_head *old_info_hash, *old_laddrhash;
1274 	unsigned int old_size = fib_info_hash_size;
1275 	unsigned int i;
1276 
1277 	spin_lock_bh(&fib_info_lock);
1278 	old_info_hash = fib_info_hash;
1279 	old_laddrhash = fib_info_laddrhash;
1280 	fib_info_hash_size = new_size;
1281 	fib_info_hash_bits = ilog2(new_size);
1282 
1283 	for (i = 0; i < old_size; i++) {
1284 		struct hlist_head *head = &fib_info_hash[i];
1285 		struct hlist_node *n;
1286 		struct fib_info *fi;
1287 
1288 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1289 			struct hlist_head *dest;
1290 			unsigned int new_hash;
1291 
1292 			new_hash = fib_info_hashfn(fi);
1293 			dest = &new_info_hash[new_hash];
1294 			hlist_add_head(&fi->fib_hash, dest);
1295 		}
1296 	}
1297 	fib_info_hash = new_info_hash;
1298 
1299 	fib_info_laddrhash = new_laddrhash;
1300 	for (i = 0; i < old_size; i++) {
1301 		struct hlist_head *lhead = &old_laddrhash[i];
1302 		struct hlist_node *n;
1303 		struct fib_info *fi;
1304 
1305 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1306 			struct hlist_head *ldest;
1307 
1308 			ldest = fib_info_laddrhash_bucket(fi->fib_net,
1309 							  fi->fib_prefsrc);
1310 			hlist_add_head(&fi->fib_lhash, ldest);
1311 		}
1312 	}
1313 
1314 	spin_unlock_bh(&fib_info_lock);
1315 
1316 	kvfree(old_info_hash);
1317 	kvfree(old_laddrhash);
1318 }
1319 
1320 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1321 				 unsigned char scope)
1322 {
1323 	struct fib_nh *nh;
1324 
1325 	if (nhc->nhc_family != AF_INET)
1326 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1327 
1328 	nh = container_of(nhc, struct fib_nh, nh_common);
1329 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1330 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1331 
1332 	return nh->nh_saddr;
1333 }
1334 
1335 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1336 {
1337 	struct fib_nh_common *nhc = res->nhc;
1338 
1339 	if (res->fi->fib_prefsrc)
1340 		return res->fi->fib_prefsrc;
1341 
1342 	if (nhc->nhc_family == AF_INET) {
1343 		struct fib_nh *nh;
1344 
1345 		nh = container_of(nhc, struct fib_nh, nh_common);
1346 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1347 			return nh->nh_saddr;
1348 	}
1349 
1350 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1351 }
1352 
1353 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1354 {
1355 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1356 	    fib_prefsrc != cfg->fc_dst) {
1357 		u32 tb_id = cfg->fc_table;
1358 		int rc;
1359 
1360 		if (tb_id == RT_TABLE_MAIN)
1361 			tb_id = RT_TABLE_LOCAL;
1362 
1363 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1364 					  fib_prefsrc, tb_id);
1365 
1366 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1367 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1368 						  fib_prefsrc, RT_TABLE_LOCAL);
1369 		}
1370 
1371 		if (rc != RTN_LOCAL)
1372 			return false;
1373 	}
1374 	return true;
1375 }
1376 
1377 struct fib_info *fib_create_info(struct fib_config *cfg,
1378 				 struct netlink_ext_ack *extack)
1379 {
1380 	int err;
1381 	struct fib_info *fi = NULL;
1382 	struct nexthop *nh = NULL;
1383 	struct fib_info *ofi;
1384 	int nhs = 1;
1385 	struct net *net = cfg->fc_nlinfo.nl_net;
1386 
1387 	if (cfg->fc_type > RTN_MAX)
1388 		goto err_inval;
1389 
1390 	/* Fast check to catch the most weird cases */
1391 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1392 		NL_SET_ERR_MSG(extack, "Invalid scope");
1393 		goto err_inval;
1394 	}
1395 
1396 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1397 		NL_SET_ERR_MSG(extack,
1398 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1399 		goto err_inval;
1400 	}
1401 
1402 	if (cfg->fc_nh_id) {
1403 		if (!cfg->fc_mx) {
1404 			fi = fib_find_info_nh(net, cfg);
1405 			if (fi) {
1406 				refcount_inc(&fi->fib_treeref);
1407 				return fi;
1408 			}
1409 		}
1410 
1411 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1412 		if (!nh) {
1413 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1414 			goto err_inval;
1415 		}
1416 		nhs = 0;
1417 	}
1418 
1419 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1420 	if (cfg->fc_mp) {
1421 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1422 		if (nhs == 0)
1423 			goto err_inval;
1424 	}
1425 #endif
1426 
1427 	err = -ENOBUFS;
1428 
1429 	/* Paired with WRITE_ONCE() in fib_release_info() */
1430 	if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1431 		unsigned int new_size = fib_info_hash_size << 1;
1432 		struct hlist_head *new_info_hash;
1433 		struct hlist_head *new_laddrhash;
1434 		size_t bytes;
1435 
1436 		if (!new_size)
1437 			new_size = 16;
1438 		bytes = (size_t)new_size * sizeof(struct hlist_head *);
1439 		new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1440 		new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1441 		if (!new_info_hash || !new_laddrhash) {
1442 			kvfree(new_info_hash);
1443 			kvfree(new_laddrhash);
1444 		} else {
1445 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1446 		}
1447 		if (!fib_info_hash_size)
1448 			goto failure;
1449 	}
1450 
1451 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1452 	if (!fi)
1453 		goto failure;
1454 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1455 					      cfg->fc_mx_len, extack);
1456 	if (IS_ERR(fi->fib_metrics)) {
1457 		err = PTR_ERR(fi->fib_metrics);
1458 		kfree(fi);
1459 		return ERR_PTR(err);
1460 	}
1461 
1462 	fi->fib_net = net;
1463 	fi->fib_protocol = cfg->fc_protocol;
1464 	fi->fib_scope = cfg->fc_scope;
1465 	fi->fib_flags = cfg->fc_flags;
1466 	fi->fib_priority = cfg->fc_priority;
1467 	fi->fib_prefsrc = cfg->fc_prefsrc;
1468 	fi->fib_type = cfg->fc_type;
1469 	fi->fib_tb_id = cfg->fc_table;
1470 
1471 	fi->fib_nhs = nhs;
1472 	if (nh) {
1473 		if (!nexthop_get(nh)) {
1474 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1475 			err = -EINVAL;
1476 		} else {
1477 			err = 0;
1478 			fi->nh = nh;
1479 		}
1480 	} else {
1481 		change_nexthops(fi) {
1482 			nexthop_nh->nh_parent = fi;
1483 		} endfor_nexthops(fi)
1484 
1485 		if (cfg->fc_mp)
1486 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1487 					  extack);
1488 		else
1489 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1490 	}
1491 
1492 	if (err != 0)
1493 		goto failure;
1494 
1495 	if (fib_props[cfg->fc_type].error) {
1496 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1497 			NL_SET_ERR_MSG(extack,
1498 				       "Gateway, device and multipath can not be specified for this route type");
1499 			goto err_inval;
1500 		}
1501 		goto link_it;
1502 	} else {
1503 		switch (cfg->fc_type) {
1504 		case RTN_UNICAST:
1505 		case RTN_LOCAL:
1506 		case RTN_BROADCAST:
1507 		case RTN_ANYCAST:
1508 		case RTN_MULTICAST:
1509 			break;
1510 		default:
1511 			NL_SET_ERR_MSG(extack, "Invalid route type");
1512 			goto err_inval;
1513 		}
1514 	}
1515 
1516 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1517 		NL_SET_ERR_MSG(extack, "Invalid scope");
1518 		goto err_inval;
1519 	}
1520 
1521 	if (fi->nh) {
1522 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1523 		if (err)
1524 			goto failure;
1525 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1526 		struct fib_nh *nh = fi->fib_nh;
1527 
1528 		/* Local address is added. */
1529 		if (nhs != 1) {
1530 			NL_SET_ERR_MSG(extack,
1531 				       "Route with host scope can not have multiple nexthops");
1532 			goto err_inval;
1533 		}
1534 		if (nh->fib_nh_gw_family) {
1535 			NL_SET_ERR_MSG(extack,
1536 				       "Route with host scope can not have a gateway");
1537 			goto err_inval;
1538 		}
1539 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1540 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1541 		err = -ENODEV;
1542 		if (!nh->fib_nh_dev)
1543 			goto failure;
1544 		netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1545 				     GFP_KERNEL);
1546 	} else {
1547 		int linkdown = 0;
1548 
1549 		change_nexthops(fi) {
1550 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1551 					   cfg->fc_table, cfg->fc_scope,
1552 					   extack);
1553 			if (err != 0)
1554 				goto failure;
1555 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1556 				linkdown++;
1557 		} endfor_nexthops(fi)
1558 		if (linkdown == fi->fib_nhs)
1559 			fi->fib_flags |= RTNH_F_LINKDOWN;
1560 	}
1561 
1562 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1563 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1564 		goto err_inval;
1565 	}
1566 
1567 	if (!fi->nh) {
1568 		change_nexthops(fi) {
1569 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1570 						  fi->fib_scope);
1571 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1572 				fi->fib_nh_is_v6 = true;
1573 		} endfor_nexthops(fi)
1574 
1575 		fib_rebalance(fi);
1576 	}
1577 
1578 link_it:
1579 	ofi = fib_find_info(fi);
1580 	if (ofi) {
1581 		fi->fib_dead = 1;
1582 		free_fib_info(fi);
1583 		refcount_inc(&ofi->fib_treeref);
1584 		return ofi;
1585 	}
1586 
1587 	refcount_set(&fi->fib_treeref, 1);
1588 	refcount_set(&fi->fib_clntref, 1);
1589 	spin_lock_bh(&fib_info_lock);
1590 	fib_info_cnt++;
1591 	hlist_add_head(&fi->fib_hash,
1592 		       &fib_info_hash[fib_info_hashfn(fi)]);
1593 	if (fi->fib_prefsrc) {
1594 		struct hlist_head *head;
1595 
1596 		head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1597 		hlist_add_head(&fi->fib_lhash, head);
1598 	}
1599 	if (fi->nh) {
1600 		list_add(&fi->nh_list, &nh->fi_list);
1601 	} else {
1602 		change_nexthops(fi) {
1603 			struct hlist_head *head;
1604 
1605 			if (!nexthop_nh->fib_nh_dev)
1606 				continue;
1607 			head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1608 			hlist_add_head(&nexthop_nh->nh_hash, head);
1609 		} endfor_nexthops(fi)
1610 	}
1611 	spin_unlock_bh(&fib_info_lock);
1612 	return fi;
1613 
1614 err_inval:
1615 	err = -EINVAL;
1616 
1617 failure:
1618 	if (fi) {
1619 		fi->fib_dead = 1;
1620 		free_fib_info(fi);
1621 	}
1622 
1623 	return ERR_PTR(err);
1624 }
1625 
1626 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1627 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1628 {
1629 	if (nhc->nhc_flags & RTNH_F_DEAD)
1630 		*flags |= RTNH_F_DEAD;
1631 
1632 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1633 		*flags |= RTNH_F_LINKDOWN;
1634 
1635 		rcu_read_lock();
1636 		switch (nhc->nhc_family) {
1637 		case AF_INET:
1638 			if (ip_ignore_linkdown(nhc->nhc_dev))
1639 				*flags |= RTNH_F_DEAD;
1640 			break;
1641 		case AF_INET6:
1642 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1643 				*flags |= RTNH_F_DEAD;
1644 			break;
1645 		}
1646 		rcu_read_unlock();
1647 	}
1648 
1649 	switch (nhc->nhc_gw_family) {
1650 	case AF_INET:
1651 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1652 			goto nla_put_failure;
1653 		break;
1654 	case AF_INET6:
1655 		/* if gateway family does not match nexthop family
1656 		 * gateway is encoded as RTA_VIA
1657 		 */
1658 		if (rt_family != nhc->nhc_gw_family) {
1659 			int alen = sizeof(struct in6_addr);
1660 			struct nlattr *nla;
1661 			struct rtvia *via;
1662 
1663 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1664 			if (!nla)
1665 				goto nla_put_failure;
1666 
1667 			via = nla_data(nla);
1668 			via->rtvia_family = AF_INET6;
1669 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1670 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1671 					    &nhc->nhc_gw.ipv6) < 0) {
1672 			goto nla_put_failure;
1673 		}
1674 		break;
1675 	}
1676 
1677 	*flags |= (nhc->nhc_flags &
1678 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1679 
1680 	if (!skip_oif && nhc->nhc_dev &&
1681 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1682 		goto nla_put_failure;
1683 
1684 	if (nhc->nhc_lwtstate &&
1685 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1686 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1687 		goto nla_put_failure;
1688 
1689 	return 0;
1690 
1691 nla_put_failure:
1692 	return -EMSGSIZE;
1693 }
1694 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1695 
1696 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1697 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1698 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1699 {
1700 	const struct net_device *dev = nhc->nhc_dev;
1701 	struct rtnexthop *rtnh;
1702 	unsigned char flags = 0;
1703 
1704 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1705 	if (!rtnh)
1706 		goto nla_put_failure;
1707 
1708 	rtnh->rtnh_hops = nh_weight - 1;
1709 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1710 
1711 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1712 		goto nla_put_failure;
1713 
1714 	rtnh->rtnh_flags = flags;
1715 
1716 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1717 		goto nla_put_failure;
1718 
1719 	/* length of rtnetlink header + attributes */
1720 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1721 
1722 	return 0;
1723 
1724 nla_put_failure:
1725 	return -EMSGSIZE;
1726 }
1727 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1728 #endif
1729 
1730 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1731 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1732 {
1733 	struct nlattr *mp;
1734 
1735 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1736 	if (!mp)
1737 		goto nla_put_failure;
1738 
1739 	if (unlikely(fi->nh)) {
1740 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1741 			goto nla_put_failure;
1742 		goto mp_end;
1743 	}
1744 
1745 	for_nexthops(fi) {
1746 		u32 nh_tclassid = 0;
1747 #ifdef CONFIG_IP_ROUTE_CLASSID
1748 		nh_tclassid = nh->nh_tclassid;
1749 #endif
1750 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1751 				    AF_INET, nh_tclassid) < 0)
1752 			goto nla_put_failure;
1753 	} endfor_nexthops(fi);
1754 
1755 mp_end:
1756 	nla_nest_end(skb, mp);
1757 
1758 	return 0;
1759 
1760 nla_put_failure:
1761 	return -EMSGSIZE;
1762 }
1763 #else
1764 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1765 {
1766 	return 0;
1767 }
1768 #endif
1769 
1770 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1771 		  const struct fib_rt_info *fri, unsigned int flags)
1772 {
1773 	unsigned int nhs = fib_info_num_path(fri->fi);
1774 	struct fib_info *fi = fri->fi;
1775 	u32 tb_id = fri->tb_id;
1776 	struct nlmsghdr *nlh;
1777 	struct rtmsg *rtm;
1778 
1779 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1780 	if (!nlh)
1781 		return -EMSGSIZE;
1782 
1783 	rtm = nlmsg_data(nlh);
1784 	rtm->rtm_family = AF_INET;
1785 	rtm->rtm_dst_len = fri->dst_len;
1786 	rtm->rtm_src_len = 0;
1787 	rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1788 	if (tb_id < 256)
1789 		rtm->rtm_table = tb_id;
1790 	else
1791 		rtm->rtm_table = RT_TABLE_COMPAT;
1792 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1793 		goto nla_put_failure;
1794 	rtm->rtm_type = fri->type;
1795 	rtm->rtm_flags = fi->fib_flags;
1796 	rtm->rtm_scope = fi->fib_scope;
1797 	rtm->rtm_protocol = fi->fib_protocol;
1798 
1799 	if (rtm->rtm_dst_len &&
1800 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1801 		goto nla_put_failure;
1802 	if (fi->fib_priority &&
1803 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1804 		goto nla_put_failure;
1805 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1806 		goto nla_put_failure;
1807 
1808 	if (fi->fib_prefsrc &&
1809 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1810 		goto nla_put_failure;
1811 
1812 	if (fi->nh) {
1813 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1814 			goto nla_put_failure;
1815 		if (nexthop_is_blackhole(fi->nh))
1816 			rtm->rtm_type = RTN_BLACKHOLE;
1817 		if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1818 			goto offload;
1819 	}
1820 
1821 	if (nhs == 1) {
1822 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1823 		unsigned char flags = 0;
1824 
1825 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1826 			goto nla_put_failure;
1827 
1828 		rtm->rtm_flags = flags;
1829 #ifdef CONFIG_IP_ROUTE_CLASSID
1830 		if (nhc->nhc_family == AF_INET) {
1831 			struct fib_nh *nh;
1832 
1833 			nh = container_of(nhc, struct fib_nh, nh_common);
1834 			if (nh->nh_tclassid &&
1835 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1836 				goto nla_put_failure;
1837 		}
1838 #endif
1839 	} else {
1840 		if (fib_add_multipath(skb, fi) < 0)
1841 			goto nla_put_failure;
1842 	}
1843 
1844 offload:
1845 	if (fri->offload)
1846 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1847 	if (fri->trap)
1848 		rtm->rtm_flags |= RTM_F_TRAP;
1849 	if (fri->offload_failed)
1850 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1851 
1852 	nlmsg_end(skb, nlh);
1853 	return 0;
1854 
1855 nla_put_failure:
1856 	nlmsg_cancel(skb, nlh);
1857 	return -EMSGSIZE;
1858 }
1859 
1860 /*
1861  * Update FIB if:
1862  * - local address disappeared -> we must delete all the entries
1863  *   referring to it.
1864  * - device went down -> we must shutdown all nexthops going via it.
1865  */
1866 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1867 {
1868 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1869 	struct net *net = dev_net(dev);
1870 	struct hlist_head *head;
1871 	struct fib_info *fi;
1872 	int ret = 0;
1873 
1874 	if (!fib_info_laddrhash || local == 0)
1875 		return 0;
1876 
1877 	head = fib_info_laddrhash_bucket(net, local);
1878 	hlist_for_each_entry(fi, head, fib_lhash) {
1879 		if (!net_eq(fi->fib_net, net) ||
1880 		    fi->fib_tb_id != tb_id)
1881 			continue;
1882 		if (fi->fib_prefsrc == local) {
1883 			fi->fib_flags |= RTNH_F_DEAD;
1884 			ret++;
1885 		}
1886 	}
1887 	return ret;
1888 }
1889 
1890 static int call_fib_nh_notifiers(struct fib_nh *nh,
1891 				 enum fib_event_type event_type)
1892 {
1893 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1894 	struct fib_nh_notifier_info info = {
1895 		.fib_nh = nh,
1896 	};
1897 
1898 	switch (event_type) {
1899 	case FIB_EVENT_NH_ADD:
1900 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1901 			break;
1902 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1903 			break;
1904 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1905 					   &info.info);
1906 	case FIB_EVENT_NH_DEL:
1907 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1908 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1909 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1910 						   event_type, &info.info);
1911 		break;
1912 	default:
1913 		break;
1914 	}
1915 
1916 	return NOTIFY_DONE;
1917 }
1918 
1919 /* Update the PMTU of exceptions when:
1920  * - the new MTU of the first hop becomes smaller than the PMTU
1921  * - the old MTU was the same as the PMTU, and it limited discovery of
1922  *   larger MTUs on the path. With that limit raised, we can now
1923  *   discover larger MTUs
1924  * A special case is locked exceptions, for which the PMTU is smaller
1925  * than the minimal accepted PMTU:
1926  * - if the new MTU is greater than the PMTU, don't make any change
1927  * - otherwise, unlock and set PMTU
1928  */
1929 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1930 {
1931 	struct fnhe_hash_bucket *bucket;
1932 	int i;
1933 
1934 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1935 	if (!bucket)
1936 		return;
1937 
1938 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1939 		struct fib_nh_exception *fnhe;
1940 
1941 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1942 		     fnhe;
1943 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1944 			if (fnhe->fnhe_mtu_locked) {
1945 				if (new <= fnhe->fnhe_pmtu) {
1946 					fnhe->fnhe_pmtu = new;
1947 					fnhe->fnhe_mtu_locked = false;
1948 				}
1949 			} else if (new < fnhe->fnhe_pmtu ||
1950 				   orig == fnhe->fnhe_pmtu) {
1951 				fnhe->fnhe_pmtu = new;
1952 			}
1953 		}
1954 	}
1955 }
1956 
1957 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1958 {
1959 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1960 	struct fib_nh *nh;
1961 
1962 	hlist_for_each_entry(nh, head, nh_hash) {
1963 		if (nh->fib_nh_dev == dev)
1964 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1965 	}
1966 }
1967 
1968 /* Event              force Flags           Description
1969  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1970  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1971  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1972  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1973  *
1974  * only used when fib_nh is built into fib_info
1975  */
1976 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1977 {
1978 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1979 	struct fib_info *prev_fi = NULL;
1980 	int scope = RT_SCOPE_NOWHERE;
1981 	struct fib_nh *nh;
1982 	int ret = 0;
1983 
1984 	if (force)
1985 		scope = -1;
1986 
1987 	hlist_for_each_entry(nh, head, nh_hash) {
1988 		struct fib_info *fi = nh->nh_parent;
1989 		int dead;
1990 
1991 		BUG_ON(!fi->fib_nhs);
1992 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1993 			continue;
1994 		prev_fi = fi;
1995 		dead = 0;
1996 		change_nexthops(fi) {
1997 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1998 				dead++;
1999 			else if (nexthop_nh->fib_nh_dev == dev &&
2000 				 nexthop_nh->fib_nh_scope != scope) {
2001 				switch (event) {
2002 				case NETDEV_DOWN:
2003 				case NETDEV_UNREGISTER:
2004 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2005 					fallthrough;
2006 				case NETDEV_CHANGE:
2007 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2008 					break;
2009 				}
2010 				call_fib_nh_notifiers(nexthop_nh,
2011 						      FIB_EVENT_NH_DEL);
2012 				dead++;
2013 			}
2014 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2015 			if (event == NETDEV_UNREGISTER &&
2016 			    nexthop_nh->fib_nh_dev == dev) {
2017 				dead = fi->fib_nhs;
2018 				break;
2019 			}
2020 #endif
2021 		} endfor_nexthops(fi)
2022 		if (dead == fi->fib_nhs) {
2023 			switch (event) {
2024 			case NETDEV_DOWN:
2025 			case NETDEV_UNREGISTER:
2026 				fi->fib_flags |= RTNH_F_DEAD;
2027 				fallthrough;
2028 			case NETDEV_CHANGE:
2029 				fi->fib_flags |= RTNH_F_LINKDOWN;
2030 				break;
2031 			}
2032 			ret++;
2033 		}
2034 
2035 		fib_rebalance(fi);
2036 	}
2037 
2038 	return ret;
2039 }
2040 
2041 /* Must be invoked inside of an RCU protected region.  */
2042 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2043 {
2044 	struct fib_info *fi = NULL, *last_resort = NULL;
2045 	struct hlist_head *fa_head = res->fa_head;
2046 	struct fib_table *tb = res->table;
2047 	u8 slen = 32 - res->prefixlen;
2048 	int order = -1, last_idx = -1;
2049 	struct fib_alias *fa, *fa1 = NULL;
2050 	u32 last_prio = res->fi->fib_priority;
2051 	dscp_t last_dscp = 0;
2052 
2053 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2054 		struct fib_info *next_fi = fa->fa_info;
2055 		struct fib_nh_common *nhc;
2056 
2057 		if (fa->fa_slen != slen)
2058 			continue;
2059 		if (fa->fa_dscp &&
2060 		    fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2061 			continue;
2062 		if (fa->tb_id != tb->tb_id)
2063 			continue;
2064 		if (next_fi->fib_priority > last_prio &&
2065 		    fa->fa_dscp == last_dscp) {
2066 			if (last_dscp)
2067 				continue;
2068 			break;
2069 		}
2070 		if (next_fi->fib_flags & RTNH_F_DEAD)
2071 			continue;
2072 		last_dscp = fa->fa_dscp;
2073 		last_prio = next_fi->fib_priority;
2074 
2075 		if (next_fi->fib_scope != res->scope ||
2076 		    fa->fa_type != RTN_UNICAST)
2077 			continue;
2078 
2079 		nhc = fib_info_nhc(next_fi, 0);
2080 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2081 			continue;
2082 
2083 		fib_alias_accessed(fa);
2084 
2085 		if (!fi) {
2086 			if (next_fi != res->fi)
2087 				break;
2088 			fa1 = fa;
2089 		} else if (!fib_detect_death(fi, order, &last_resort,
2090 					     &last_idx, fa1->fa_default)) {
2091 			fib_result_assign(res, fi);
2092 			fa1->fa_default = order;
2093 			goto out;
2094 		}
2095 		fi = next_fi;
2096 		order++;
2097 	}
2098 
2099 	if (order <= 0 || !fi) {
2100 		if (fa1)
2101 			fa1->fa_default = -1;
2102 		goto out;
2103 	}
2104 
2105 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2106 			      fa1->fa_default)) {
2107 		fib_result_assign(res, fi);
2108 		fa1->fa_default = order;
2109 		goto out;
2110 	}
2111 
2112 	if (last_idx >= 0)
2113 		fib_result_assign(res, last_resort);
2114 	fa1->fa_default = last_idx;
2115 out:
2116 	return;
2117 }
2118 
2119 /*
2120  * Dead device goes up. We wake up dead nexthops.
2121  * It takes sense only on multipath routes.
2122  *
2123  * only used when fib_nh is built into fib_info
2124  */
2125 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2126 {
2127 	struct fib_info *prev_fi;
2128 	struct hlist_head *head;
2129 	struct fib_nh *nh;
2130 	int ret;
2131 
2132 	if (!(dev->flags & IFF_UP))
2133 		return 0;
2134 
2135 	if (nh_flags & RTNH_F_DEAD) {
2136 		unsigned int flags = dev_get_flags(dev);
2137 
2138 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2139 			nh_flags |= RTNH_F_LINKDOWN;
2140 	}
2141 
2142 	prev_fi = NULL;
2143 	head = fib_info_devhash_bucket(dev);
2144 	ret = 0;
2145 
2146 	hlist_for_each_entry(nh, head, nh_hash) {
2147 		struct fib_info *fi = nh->nh_parent;
2148 		int alive;
2149 
2150 		BUG_ON(!fi->fib_nhs);
2151 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2152 			continue;
2153 
2154 		prev_fi = fi;
2155 		alive = 0;
2156 		change_nexthops(fi) {
2157 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2158 				alive++;
2159 				continue;
2160 			}
2161 			if (!nexthop_nh->fib_nh_dev ||
2162 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2163 				continue;
2164 			if (nexthop_nh->fib_nh_dev != dev ||
2165 			    !__in_dev_get_rtnl(dev))
2166 				continue;
2167 			alive++;
2168 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2169 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2170 		} endfor_nexthops(fi)
2171 
2172 		if (alive > 0) {
2173 			fi->fib_flags &= ~nh_flags;
2174 			ret++;
2175 		}
2176 
2177 		fib_rebalance(fi);
2178 	}
2179 
2180 	return ret;
2181 }
2182 
2183 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2184 static bool fib_good_nh(const struct fib_nh *nh)
2185 {
2186 	int state = NUD_REACHABLE;
2187 
2188 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2189 		struct neighbour *n;
2190 
2191 		rcu_read_lock_bh();
2192 
2193 		if (likely(nh->fib_nh_gw_family == AF_INET))
2194 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2195 						   (__force u32)nh->fib_nh_gw4);
2196 		else if (nh->fib_nh_gw_family == AF_INET6)
2197 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2198 							   &nh->fib_nh_gw6);
2199 		else
2200 			n = NULL;
2201 		if (n)
2202 			state = n->nud_state;
2203 
2204 		rcu_read_unlock_bh();
2205 	}
2206 
2207 	return !!(state & NUD_VALID);
2208 }
2209 
2210 void fib_select_multipath(struct fib_result *res, int hash)
2211 {
2212 	struct fib_info *fi = res->fi;
2213 	struct net *net = fi->fib_net;
2214 	bool first = false;
2215 
2216 	if (unlikely(res->fi->nh)) {
2217 		nexthop_path_fib_result(res, hash);
2218 		return;
2219 	}
2220 
2221 	change_nexthops(fi) {
2222 		if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2223 			if (!fib_good_nh(nexthop_nh))
2224 				continue;
2225 			if (!first) {
2226 				res->nh_sel = nhsel;
2227 				res->nhc = &nexthop_nh->nh_common;
2228 				first = true;
2229 			}
2230 		}
2231 
2232 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2233 			continue;
2234 
2235 		res->nh_sel = nhsel;
2236 		res->nhc = &nexthop_nh->nh_common;
2237 		return;
2238 	} endfor_nexthops(fi);
2239 }
2240 #endif
2241 
2242 void fib_select_path(struct net *net, struct fib_result *res,
2243 		     struct flowi4 *fl4, const struct sk_buff *skb)
2244 {
2245 	if (fl4->flowi4_oif)
2246 		goto check_saddr;
2247 
2248 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2249 	if (fib_info_num_path(res->fi) > 1) {
2250 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2251 
2252 		fib_select_multipath(res, h);
2253 	}
2254 	else
2255 #endif
2256 	if (!res->prefixlen &&
2257 	    res->table->tb_num_default > 1 &&
2258 	    res->type == RTN_UNICAST)
2259 		fib_select_default(fl4, res);
2260 
2261 check_saddr:
2262 	if (!fl4->saddr)
2263 		fl4->saddr = fib_result_prefsrc(net, res);
2264 }
2265